Methods and apparatus for end-to-end document image quality assessment using machine learning without having ground truth for characters

2. The apparatus of claim 1, wherein the quality assessment machine learning model includes an end-to-end machine learning model.

3. The apparatus of claim 1, wherein the quality assessment machine learning model is included in a three-dimensional convolutional neural network.

4. The apparatus of claim 1, wherein the instructions to cause the processor to process the plurality of image patches to calculate the quality target score include instructions to further cause the processor to calculate the quality target score of the plurality of quality target scores for each image patch of the plurality of image patches based on a number of correct natural language readable data and the natural language control source.

5. The apparatus of claim 1, wherein the instructions to cause the processor to train the quality assessment machine learning model includes instructions to further cause the processor to train the quality assessment machine learning model using a training set, the training set including readable data correlated to a target score.

6. The apparatus of claim 5, wherein the instructions to cause the processor to train the quality assessment machine learning model includes instructions to further cause the processor to train the quality assessment machine learning model without labeling the training set.

7. The apparatus of claim 1, wherein the memory stores instructions to further cause the processor to reduce a size of each image patch from the plurality of image patches, to increase a number of image patches in the patch filter window, prior to processing the plurality of image patches to calculate the quality target score of the plurality of quality target scores.

8. The apparatus of claim 1, wherein the memory stores instructions to further cause the processor to adjust a size of each image patch from the plurality of image patches based on a character threshold, prior to processing the plurality of image patches to calculate the quality target score of the plurality of quality target scores, the character threshold indicating a minimum count of readable data in each image patch of the plurality of image patches.

10. The apparatus of claim 7, wherein the memory stores instructions to cause the processor to update the quality target score of one or more overlapping image patches between the first cluster of image patches and the second cluster of image patches.

11. The apparatus of claim 1, wherein the instructions to cause the processor to train the quality assessment machine learning model includes instructions to further cause the processor to train the quality assessment machine learning model automatically.

13. The apparatus of claim 12, wherein the memory stores instructions to further cause the processor to iteratively train the trained qualitative assessment machine learning model using future document image data.

14. The apparatus of claim 13, wherein the instructions to cause the processor to generate the qualitative value score for the document image data includes instructions to further cause the processor to generate the qualitative value score based on the plurality of quality target scores concurrently during iterative training of the trained quality assessment machine learning model using the future document image data.

19. The method of claim 18, wherein each of receiving the first document image, processing, training, receiving the second document image, extracting, executing and generating is performed automatically.

21. The method of claim 20, wherein the calculating the second cluster of quality target scores for each image patch from the second cluster of image patches includes updating the quality target score of one or more overlapping image patches between the first cluster of image patches and the second cluster of image patches.